Background: Human Pose Estimation (HPE) has gained increasing attention in sports research due to advancements in Deep Learning (DL) movement skills, which enable precise joint localization in 2D and 3D visual data. DL-based HPE facilitates non-invasive analysis of movement patterns in real-world settings, providing actionable insights for training, performance optimisation, and injury prevention. This systematic review examines the application of DL-based HPE in sports, focusing on the availability and accessibility of training datasets, reproducibility for practitioners, and the influence of human factors. The review also offers recommendations to guide future research and applications. Methods: A systematic search following PRISMA guidelines was conducted across four databases—Scopus, Web of Science, the Association for Computing Machinery, and SPORTDiscus, yielding 371 articles. Two independent reviewers applied inclusion and exclusion criteria to identify relevant studies, with a third reviewer resolving conflicts. Key aspects analysed included the scope of DL-based HPE applications, dataset characteristics, and algorithmic approaches. A supplementary search was conducted to include contemporary literature published since the initial search date. Data were synthesized descriptively, focusing on trends and limitations in the evidence base. Results: The identified applications of DL-based HPE in sports were categorized into four domains: movement skill analysis, action recognition, augmented coaching tools, and officiating support. Most studies relied on private datasets for algorithm training and validation, limiting reproducibility and generalizability. Bespoke multi-model algorithms were the most common approach, and single person pose estimation predominated. Despite its potential, the lack of open datasets and standardized practices poses challenges for broader adoption and practical implementation. These findings were echoed in the supplementary search which added no significant findings outside what previous studies had demonstrated. Conclusions: This review represents the first systematic evaluation of DL-based HPE from a sports science perspective, offering practical guidance for future research and applications. The findings highlight the need for open, standardized datasets and reproducible methodologies to advance the field. Future research should address these limitations while exploring innovative applications to maximize the impact of DL-based HPE in sports science.

KinectFusion is a typical three-dimensional reconstruction technique which enables generation of individual three-dimensional human models from consumer depth cameras for understanding body shapes. The aim of this study was to compare three-dimensional reconstruction results obtained using KinectFusion from data collected with two different types of depth camera (time-of-flight and stereoscopic cameras) and compare these results with those of a commercial three-dimensional scanning system to determine which type of depth camera gives improved reconstruction. Torso mannequins and machined aluminium cylinders were used as the test objects for this study. Two depth cameras, Microsoft Kinect V2 and Intel Realsense D435, were selected as the representatives of time-of-flight and stereoscopic cameras, respectively, to capture scan data for the reconstruction of three-dimensional point clouds by KinectFusion techniques. The results showed that both time-of-flight and stereoscopic cameras, using the developed rotating camera rig, provided repeatable body scanning data with minimal operator-induced error. However, the time-of-flight camera generated more accurate three-dimensional point clouds than the stereoscopic sensor. Thus, this suggests that applications requiring the generation of accurate three-dimensional human models by KinectFusion techniques should consider using a time-of-flight camera, such as the Microsoft Kinect V2, as the image capturing sensor.

Somatotype is an approach to quantify body physique (shape and body composition). Somatotyping by manual measurement (the anthropometric method) or visual rating (the photoscopic method) needs technical expertize to minimize intra‐ and inter‐observer errors. This study aims to develop machine learning models which enable automatic estimation of Heath‐Carter somatotypes using a single‐camera 3D scanning system. Single‐camera 3D scanning was used to obtain 3D imaging data and computer vision techniques to extract features of body shape. Machine learning models were developed to predict participants' somatotypes from the extracted shape features. These predicted somatotypes were compared against manual measurement procedures. Data were collected from 46 participants and used as the training/validation set for model developing, whilst data collected from 17 participants were used as the test set for model evaluation. Evaluation tests showed that the 3D scanning methods enable accurate (mean error < 0.5; intraclass correlation coefficients >0.8) and precise (test‐retest root mean square error < 0.5; intraclass correlation coefficients >0.8) somatotype predictions. This study shows that the 3D scanning methods could be used as an alternative to traditional somatotyping approaches after the current models improve with the large datasets.

The purpose of this study was to determine the kinematic patterns that maximized the vertical force produced during the water polo eggbeater kick. Twelve water polo players were tested executing the eggbeater kick with the trunk aligned vertically and with the upper limbs above water while trying to maintain as high a position as possible out of the water for nine eggbeater kick cycles. Lower limb joint angular kinematics, pitch angles and speed of the feet were calculated. The vertical force produced during the eggbeater kick cycle was calculated using inverse dynamics for the independent lower body segments and combined upper body segments, and a participant-specific second-degree regression equation for the weight and buoyancy contributions. Vertical force normalized to body weight was associated with hip flexion (average, r = 0.691; maximum, r = 0.791; range of motion, r = 0.710), hip abduction (maximum, r = 0.654), knee flexion (average, r = 0.716; minimum, r = 0.653) and knee flexion-extension angular velocity (r = 0.758). Effective orientation of the hips resulted in fast horizontal motion of the feet with positive pitch angles. Vertical motion of the feet was negatively associated with vertical force. A multiple regression model comprising the non-collinear variables of maximum hip abduction, hip flexion range of motion and knee flexion angular velocity accounted for 81% of the variance in normalized vertical force. For high performance in the water polo, eggbeater kick players should execute fast horizontal motion with the feet by having large abduction and flexion of the hips, and fast extension and flexion of the knees.

The ability to meet high exercise intensities is limited by the increased risk of injury in some clinical populations. Previous studies have linked large tibia peak positive accelerations resulting from running to increased risk of developing lower-extremity injury. The purpose of this study is to determine the feasibility of using a hip flexion feedback system (HFFS) to meet and maintain different exercise intensities while maintaining low tibia axial accelerations. Ten healthy participants were tested on a HFFS test and an independent walking/running test to meet exercise intensities of 40% and 60% of heart rate reserve (HRR). During the HFFS test, the HFFS controlled in real time the exercise intensity by directing individuals to specific maximum hip flexion targets during walking and providing visual information that assists them in maintaining low tibia peak positive accelerations during the initial contact phase. Maximum hip flexion targets during walking are calculated based on real-time readings of the participant’s heart rate. During the independent test, exercise intensity was controlled independently by the participant using treadmill speed. Compared to the independent test, using the HFFS at 60% HRR resulted in similar heart-rate error but lower tibia peak positive accelerations. No differences were observed for the 40% HRR intensity. This paper describes a novel exercise approach that uses the individual’s heart rate to calculate maximal hip flexion targets that an individual should meet during treadmill walking. The HFFS also provides tibia peak positive peak acceleration cues. Therefore, the HFFS can increase and control exercise intensities while maintaining low tibia accelerations. In particular, the HFFS might be an alternative strategy to meet moderate to vigorous exercise intensities in populations at risk of developing lower-extremity injuries.